Anduril Swaps Drone's AI Brain Mid-Flight
In a test for the U.S. Air Force's Collaborative Combat Aircraft (CCA) program, an Anduril drone successfully swapped its AI-powered flight control software while airborne. The YFQ-44A drone transitioned from a system developed by Shield AI to Anduril's own Lattice autonomy stack, demonstrating a first-of-its-kind modular software capability. The Air Force plans to select a winner for the CCA program, which pairs autonomous drones with manned fighters, by the end of 2026.
This successful AI software swap was enabled by the Air Force's Autonomy Government Reference Architecture (A-GRA). This standard intentionally decouples the flight-critical control software from the mission autonomy software, creating a standardized interface that allows different "AI brains" to be plugged into an aircraft without affecting its airworthiness certification. The test aircraft, Anduril's YFQ-44A, is a 20-foot-long drone powered by a Williams FJ44-4M turbofan engine, capable of reaching speeds of Mach 0.95 and sustaining 9g maneuvers. The platform, originally developed by Blue Force Technologies before its acquisition by Anduril, recently began captive-carry flight tests with inert AIM-120 AMRAAM missiles to validate weapons integration. Anduril's Lattice for Mission Autonomy is a hardware-agnostic software that fuses data from various sensors to enable teams of robotic assets to perform complex missions under human supervision. The other software tested, Shield AI's Hivemind, is an AI pilot that has already been used to autonomously fly other military aircraft, including the F-16 and MQ-20 Avenger, and is designed for operations in GPS-denied environments. This demonstration is a key validation of the Department of Defense's push for a Modular Open Systems Approach (MOSA). Standards under this approach, like the Open Mission Systems (OMS) architecture, are meant to prevent vendor lock-in by enforcing common interfaces, allowing the military to rapidly integrate the best available technology from any supplier. A significant hurdle for deploying AI in these systems is certification under standards like DO-178C, the benchmark for safety-critical airborne software. DO-178C is built on the principle of determinism—where the same input always produces the same output—a concept fundamentally at odds with the non-deterministic nature of machine learning algorithms. The choice of onboard processing hardware for such AI systems involves a trade-off between Field-Programmable Gate Arrays (FPGAs) and Graphics Processing Units (
